Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
Integrated design-to-control approach for holonic manufacturing systems Lihui Wang* National Research Council of Canada, Integrated Manufacturing Technologies Institute, 800 Collip Circle, London, ON N6G 4X8 Canada
Abstract Next generation manufacturing systems will be integrated networks of distributed resources simultaneously capable of combined knowledge and material processing. These manufacturing systems will be required to be agile, #exible, and fault-tolerant. The objective of this research is to de"ne a generic open architecture for such kind of distributed manufacturing systems, especially for holonic manufacturing systems (HMS). This paper will address issues associated with HMS, and propose a reference architecture based on a design-to-control concept. The primary focus will be given to the collaborative and integrated design-to-control approach based on machining feature, agent technology, and function block standards. Emphasis is also extended and given to metamorphic process planning and control of HMS using multi-agent negotiation and cooperation. The proposed approach, together with the open architecture, shows much promise for improving the entire manufacturing system performance under the ever-changing real-time and distributed shop #oor environments. 2001 Elsevier Science Ltd. All rights reserved. Keywords: HMS; Machining feature; Agent technology; Function block; Intelligent control
1. Introduction The era of today's design and manufacturing systems with its downstream information #ow is passing and will be gradually phased out, because the conventional design and manufacturing technologies are insu$ciently #exible, due in part to their rigid system architecture. Today's computer-aided design and manufacturing (CAD/CAM) technologies have existed for more than two decades. Despite all the past accomplishments, they have limited future potential. Further improvements can only be realized by a paradigm shift that would allow manufacturing restrictions to be considered at the early product design stage. Future collaborative design and intelligent manufacturing systems will be required to be agile, #exible, open, autonomous, and fault-tolerant. This type of system will contain collaborative and integrated design-to-control entities that can dynamically collaborate to satisfy both local and global objectives in four-dimensional space. Within the 4D space, related design-to-control information is exchanged from stream
* Corresponding author. Tel.: #1-519-430-7084; fax: #1-519-4307064. E-mail address:
[email protected] (L. Wang).
to space in all directions seamlessly, concurrently, e$ciently, and accurately. In the last decade, the change in market requirements towards a larger variety of products in smaller batch sizes, has led to the concept of next generation intelligent manufacturing systems (IMS) being an integrated network of distributed resources simultaneously capable of combined knowledge and material processing. The IMS, in the near future, will become a multi-agent system implemented within a dynamically recon"gurable factory having a decentralized and virtual organization structure. Earlier research in the area of IMS has established that such distributed resources can be realized through the concepts of the holonic paradigm [1}3], which shows promise of improving system performance and #exibility under ever-changing real-time distributed manufacturing environments. The objective of this research is to "nd a suitable and generic open architecture for distributed manufacturing systems, especially for holonic manufacturing systems (HMS). The primary focus will be given to a collaborative and integrated design-to-control approach. This paper will "rst address issues associated with HMS, and then discuss three useful implementation techniques } machining feature, agent technology, and function block standard. An integrated design-to-control architecture for the next generation HMS is proposed based
0736-5845/01/$ - see front matter 2001 Elsevier Science Ltd. All rights reserved. PII: S 0 7 3 6 - 5 8 4 5 ( 0 0 ) 0 0 0 5 0 - 8
160
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
on these enabling technologies. Emphasis will also be extended and given to metamorphic process planning and control of HMS using multi-agent negotiation and cooperation.
2. Holonic manufacturing system Based on the holonic concept [1], the next generation of IMS could form distributed re-con"gurable virtual factories, in which human, machines, and control modules interact in dynamically formed virtual clusters. Such a system could be built from intelligent, autonomous, and cooperative elements or holons [2}4]. A holon is de"ned as an autonomous and cooperative building block, within a manufacturing system, for transforming, transporting, storing and/or validating information and physical objects [3]. A holon has the autonomy to create and control the execution of its own plans, and can cooperate with other holons to develop mutually acceptable plans for achieving system goals. Cooperation among holons is accomplished through an evolutionary self-organizing holarchy (a system of holons). Such a holarchy, which integrates the entire range of manufacturing activities, such as order booking, design, production, marketing, etc., is de"ned as HMS. Fig. 1 illustrates the major functional elements and interfaces of a holon. A holon exchanges information, material, or resources with other holons via its interfaces through negotiation and cooperation. It can also form a part of another holon. Possessing autonomy, holons can not only generate their own task plans but also execute and monitor these plans while avoiding malfunctions. The cooperation among holons enables them to perform plan generation and execution via information exchange (communication) and group decision-making (negotiation). Each holon participates in one or more cooperation domains. An HMS consists of a number of such autonomous, self-reliant manufacturing units * holons. Any unit, such as a machine, a conveyor, a workpiece, or an order, can be a holon as long as the unit is able to create and control the execution of its own plans and/or strategies. A holon always contains an information processing part and optionally a physical processing part. A scheduler is an example of a holon without a physical processing part. In an HMS, the constitution of holons changes and the functional elements evolve over time according to system level requirements. For instance, a turning center holon may be augmented with certain milling operations or a holonic robot may be provided with additional vision system capabilities. Hence, the ability to recon"gure on demand is an important requirement for holonic systems. This and related capabilities of holonic systems will largely be dependent on their design, planning, and control systems.
Fig. 1. Holonic elements and interfaces.
Fig. 2 illustrates an open, generic system architecture of next generation HMS. The proposed architecture consists of three major modules (design, planning, and control) and a shared dynamic database. Each module can be considered as a holon. The functionality of these modules is enabled by inter-holon negotiation and cooperation through their associated interfaces, and by interactions with physical holons (the intelligent machine controllers, sensors, actuators, etc.). If the holons are connected by a real-time network, the concept of designto-control integration can hence be realized through communication and collaboration whenever manufacturablity checks are required.
3. Machining feature-based design in HMS From design to manufacturing of a mechanical product, a number of steps must be followed, such as geometry design, process planning, tool selection, operation optimization, NC code generation, as well as "xture design and preparation. Engineers will have to meet and respond to the challenges for producing highly complex products in very small batches featured by HMSs. As every mechanical product designed using CAD systems will need to be manufactured afterward, it is important to take downstream manufacturing/control restrictions/constraints into account in the geometry design phase, to simplify and facilitate all subsequent manufacturing processes. Conventional design activities have been practiced with downstream information #ow. Information feedback from the low-level shop #oors to the high-level designs is usually performed by human interactions. This may cause an insu$cient design and hence ine$cient product development due to the absence of manufacturability check at higher levels, based on the available resources and technologies. The concept of design-to-control is to bring design and manufacturing control issues together with high integration and with as
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
161
Fig. 2. Reference architecture of HMS.
much concurrency as possible. This fact is of vital importance for reducing the risk of designing a product which is later found to be uneconomical, di$cult or even impossible to fabricate with the available resources. A machining feature-based approach is adopted in this research to solve these problems in holonic manufacturing environment. 3.1. Machining feature-based approach The concept of machining features was introduced to product design in the early-1990s [5}7]. As summarized in Fig. 3, the machining features are the geometrical shapes, such as step, slot, pocket, and hole, etc., which can be easily achieved by the available resources and the de"ned technologies. Di!erent from design features (e.g. vertex, curve, face, etc.), each machining feature holds a set of loosely coupled information on how to fabricate it, such as cutting tool ID, machining sequence, tool-path generation logic, cutting conditions, etc. Consequently, the time and e!orts used for subsequent processes (e.g. process planning, tool selection, operation optimization, NC code generation, etc.) can be signi"cantly reduced. Designing a product by combining the machining features properly, a designer or engineer can easily assure its manufacturability. 3.2. Information yow in feature-based systems Product design is the "rst step of the design-to-control process, that deals with the conceptualization and planning of the functional and physical characteristics of a product (e.g. conceptual design, functional design, and detailed design). An ideal design plan should always address the ease of machining and assembly [8]. In other
Fig. 3. Typical examples of machining features.
words, the purpose of product design is to study how to design a product that can not only reduce the cost and time to fabricate it, but also preserve the manufacturability of the product. This is highly dependent on the information #ow of the system. Fig. 4 shows the information #ow between the design module and the other system modules. A set of resource databases including machining feature, machine, tool, and machining technology are shared among design, planning, and control modules. The generated NC code should easily be downloaded to or uploaded from any machine controllers via an execution control module. A good infrastructure should always
162
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
low-level operation planning. The former is mainly responsible for machine selection, jig/"xture selection, and machining sequence planning for a product. The latter, on the other hand, deals mainly with detailed working steps of machining operations for each machining features, including tool selection, machining strategy selection, cutting condition generation, tool path planning, and NC code generation. Fig. 5 shows the entire hierarchy of the two-level process planning. At the supervisory planning level, the outputs of the system are a set of sequenced process plans, each of which is for a single setup * one "xturing on one machine. At the operation planning level, function blocks are generated as system output, each of which describes the detailed machining operations for a single process * one tool under one set of cutting conditions. Once the machining sequence (process plan) for a part or product is decided, its detailed operation plans can be "nalized for each machining feature through multi-agent negotiation. The "nal control codes are encapsulated into a set of function blocks for distribution and execution. The details of function blocks are explained in Section 5. Fig. 4. System-wide information #ow.
4.2. Agent-based architecture for decision-making facilitate information sharing and data exchange to support decision-making at di!erent levels. As explained in the subsequent sections, function blocks and agents are used for this purpose, in addition to the machining features being major information carriers.
4. Agent-based process planning in HMS Current manufacturing technologies still require a number of preparation tasks, including "xture preparation, cutting tool preparation, and NC data preparation, to be completed before unmanned machining can take place. The necessity for these preparatory operations occurs frequently in low-repetitive workshops. Although great e!orts have been expended on feature recognition [9}12] and cutting tool selection [13}16], process planning remains as the most di$cult part of post-design processing, due to the fact of insu$cient data standardization and too many extensions of NC controllers from venders. Especially in HMS, the process planning requires large amount of negotiation for decision-making, due to the distributed nature of HMS. An agent-based approach is, therefore, selected for the decision-making in process planning for HMS. 4.1. Two-level process planning The agent-based process planning in HMS can be divided into high-level supervisory planning and
The centralized planning and control that have de"ned the traditional information processing structure of manufacturing systems is no longer suited to the rapidly changing HMS. For e$cient use of manufacturing resources and increased #exibility, it is necessary to migrate to a distributed information processing system in which entities work cooperatively towards overall system goals. This requires distributed parallel computation, asynchronous process coordination, and standard communication protocol. Such kind of planning system can be readily realized through the multi-agent paradigm. A multi-agent system usually consists of a set of autonomous interacting software entities called intelligent agents, which are knowledgeable in their local domain and share the responsibility of achieving multi-objective system goals through negotiation. The multi-agent system will contain a population of heterogeneous agents inter-operating asynchronously with other agents through collaboration and negotiation, based on a well de"ned message passing mechanism [17] in a distributed environment such as HMS. As planning task develops, the relevant agents will be dynamically grouped into coordination clusters to facilitate their focus on the current task. These clusters will be active as long as required and be destroyed when no longer needed. Fig. 6 illustrates a multi-agent architecture for process planning. As shown in the "gure, relevant agents for a particular task form, or are formed, into virtual clusters. These clusters are coordinated and arbitrated by mediator agents, such as product , planning , and resource
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
163
Fig. 5. Two-level process planning.
Fig. 6. Agent-based architecture for decision-making.
managers. These mediators are coordinator agents that possess the organizational knowledge and meta-level rules to facilitate cooperation among intra and inter agent communities. Process planning is thus conducted by the dynamic groupings of well-organized agent clusters and coordinated by the mediators. Messages are exchanged among agents within/between the clusters for communication. Based on this architecture, a process plan can be created by the collaboration of multiple cross-functional agents through negotiation and information sharing. Based on the architecture, each machine on a shop #oor is represented by an autonomous machine agent having knowledge about its own capability, assigned tooling, coolant, and spindle/axis information. Each cutting tool is represented by an autonomous tool agent which has knowledge about its tool type, dimension, remaining tool life, and tool conditions (tool wear, etc.). All the agents involved in a given task communicate with
each other asynchronously and simultaneously for local and global decision-making. For example, the feature agents cooperate with database agent and indirectly get access to feature knowledge base for feature interaction and manufacturability assessment, while tool path generating agent negotiates with machine agents, tool agents, strategy planning agent, and machining technology agent to "nd out an optimal cutter trajectory. On the other hand, the sequence planning agent is responsible for arranging an appropriate sequence of machining operations. Once the control data is "nalized by NC data generating agent, it will be passed onto the execution control agent, which interact with the real control world, for distribution and execution. All the agents use knowledge query manipulation language (KQML) as standard communication protocol. On behalf of their counterparts * physical holons, agents are utilized to represent the holons for negotiation and decision-making in the software domain.
164
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
Fig. 7. Holonic distributed control.
5. Function block-based control in HMS In an HMS, the objective is to achieve a complete spectrum of manufacturing control functions ranging from production planning at the highest level, to process/machine control at the lowest level. Whereas an HMS is inherently distributed and metamorphic, its control system generally requires numerous changes in form, substance, and allocation to be accommodated within system lifetime. The requirements for the metamorphic control in HMS are the unique combination and interaction of those of real time control, distributed control, event-driven control, and intelligent control. As shown in Fig. 7, a typical control system of this nature makes use of a real-time network for communication among spatially distributed controller nodes. The centralized application level control program that previously would have been executed on a single CPU is now distributed among the controller nodes. The distributed event-driven programs synchronize by communication through message passing under-real-time constraints to meet application requirements. 5.1. Machining process encapsulation Since the distributed CNC machine controllers are required to be intelligent enough to automatically capture the machining know-how information and recon"gure task assignments based on the ever-changing HMS environments, its control function designer, or process planner, should have the same characteristics. For the ease of control code reuse, distribution, and dynamic system recon"guration, feature-based machining process data are better to be encapsulated. This can be realized by utilizing IEC-1499 function block architecture [18]. The function block architecture is an emerging standard for distributed industrial process measurement and control systems (IPMCS). It uses an explicit event driven model and also provides for data #ow and "nite state automata-based controls which is the interest of distributed NC machining in the future HMS enterprise. Being
Fig. 8. Basic (left) and composite (right) function blocks.
an atomic distributable and executable control function unit, a function block instance encapsulates a part of machining process data (e.g. slot roughing, pocket "nishing, and hole drilling, etc.) of a given machining feature. It comprises of an individual, named copy of the data structure speci"ed by its function block class, which persists from one invocation of the function block to the next. Fig. 8 shows the de"nitions of both the basic and composite function blocks. A function block, especially the basic function block, may have multiple outputs and can maintain their internal hidden state information. In other words, a function block can generate di!erent outputs even if the same inputs are applied. This fact is of vital importance for the automatic cutting condition correction after an NC program has been downloaded to a CNC controller by changing the internal state of a function block (a machining process). For example, an NC program of the same pocket roughing can be shared by two di!erent milling machines with di!erent cutting conditions, simply by adjusting the internal state variable of the function block instants. The event #ow determines the scheduling and execution of machining operations speci"ed by the algorithms in basic function blocks. It can also send signals to maintain or change the internal state variables. Control architectures for HMS often need to meet a number of requirements, such as autonomy, reliability, fault-tolerance, interoperability, re-con"gurability, and other real-time functionality. For NC machine control, the function block is a well-established concept for de"ning robust, reusable software components with detailed machining process encapsulated. 5.2. NC code generation Before a machining operation is dispatched to an NC machine, its control code must be generated. The procedure is straightforward based on the proposed architecture. Fig. 9 shows how an NC code is generated and
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
165
Fig. 9. NC code generation and encapsulation.
Fig. 10. Function block-based control application.
encapsulated through an example of four-side pocket. As mentioned previously, a product is designed by the combination of a set of machining features. Each machining feature possesses a key parameter together with several other non-dominant parameters for its geometry de"nition. In the case of pocket, D4 serves as the key parameter, which determines the diameter of its cutting tool. Each feature type also has a set of technological information loosely coupled with it. This includes recommended tool type and tool path generation logic. In this example, a square endmill T 030 is selected for rough milling based on D4. If the material of the designed product is known, the cutting conditions (cutting speed, feed rate, depth of cut, and step over, etc.) can easily be retrieved from an
associated cutting tool database. Accordingly, the tool path for rough milling can be planned based on the tool data and cutting strategy, using pre-de"ned tool path generation logic. Once the cutting conditions and tool path are found, a partial NC code, for rough milling in this case, is generated. The piece of code is then encapsulated by appropriate function block as part of its algorithms for execution. 5.3. Function block-based control A function block-based NC control application consists of a network, in which nodes are function blocks and branches are data and event connections. Fig. 10 shows
166
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
Fig. 11. Distribution of NC control applications.
Fig. 12. Execution control mechanism.
how a function block-based application is generated for the case of pocket-milling. Typically, the necessary milling information can be extracted from meta working steps. Each meta working step (WS) is an abstractive machining process containing machining data in hierarchy, so that its instance WS can easily inherit these attributes as needed. WSs are "nally wrapped by function blocks with event and data #ows among them. The event #ow determines the scheduling and execution of NC operations speci"ed by individual WS. Events and data always #ow into the left side of each function block, while generated data and events are depicted on the right side. The well-connected function blocks are ready for distribution to machine controllers. Machining process encapsulation enables and facilitates transparent distribution of NC codes over a set of distributed machine controllers. As shown in Fig. 11, an NC program may reside completely in a single machine controller or may be distributed among several devices to accomplish the entire machining operation. The communication between function blocks can be realized by a well-established message passing mechanism under real-time constraints. This capability and others are crucial for the next generation of NC machining in the ever-changing HMS environment.
In addition to encapsulation of machining processes and communication between function blocks, appropriate execution control charts (ECC) are used to control their internal algorithms. Fig. 12 shows one typical example of ECC for pocket roughing. Being a "nite state machine, an ECC is made of EC states, EC transitions, and EC actions. The initial EC state, START in this example, cannot have any EC actions associated with it. The occurrence of an event input, such as PI}Init and PI}Cut, causes the ECC to be invoked and the input variables (toolC, key}para, etc.) to be mapped. The EC transitions use a Boolean combination of conditions that may be comprised of event inputs, input variables, output variables, and internal variables. A triggered EC transition causes a change of EC state and this leads to the execution of an associated EC action, Init or Cut in this case. The EC action then sends out an event, PI}Init or PI}Cut, upon completion. In the near future, controllers will either provide function blocks as part of their device "rmware or provide function block libraries from which function blocks can be selected and downloaded [19]. Any vender or user extensions to the standard NC code can be encapsulated by function blocks and distributed to end users. Further details regarding machining feature, agent, and function
L. Wang / Robotics and Computer Integrated Manufacturing 17 (2001) 159}167
block could be found from previous research works of L. Wang [20,21]. 6. Conclusions This paper proposed an integrated design-to-control approach for holonic manufacturing systems. It covers topics from product design, process planning, to manufacturing control, based on concepts of machining feature, agent negotiation, and function block. Architectures of two-level process planning as well as agent-based decision-making are used in addition to the reference architecture. Function blocks are recommended as new NC controller language in the future. The proposed methodology shows promise of increasing productivity and quality of product development for next generation of holonic manufacturing systems under ever-changing shop #oor environment.
References [1] Koestler A. The ghost in the machine. 1971, ISBN 0-14-019192-5. [2] Christensen JH, Norrie DH, Schae!er C. Material handling requirements in holonic manufacturing systems. Proceedings of second International Conference on Material Handling Research. Michigan, 1994. p. 1}22. [3] Christensen JH. Holonic manufacturing systems: initial architecture and standards directions. Proceedings of the First European Conference on Holonic Manufacturing Systems. Germany, 1994. p. 1}20. [4] Brennan RW, Balasubramanian S, Norrie DH. A dynamic control architecture for metamorphic control of advanced manufacturing systems. Proceedings of Intelligent Systems & Automated Manufacturing. 1997. p. 213}23. [5] Hanada T, Hoshi T. Block-like component CAD/CAM system for fully automated CAM processing. Ann CIRP 1992; 41(1):551}6.
167
[6] Sahir Arikan MA, Totuk OH. Design by using machining operations, Ann CIRP 1992;41(1):185}8. [7] Wang L, Zhao W, Ma'ruf A, Hoshi T. Setup-less fabrication technology incorporated with machining feature-based CAD/ CAM system for low volume and high product-mix machining center workshop. Proceedings of the IMEC'96. 1996. p. 95}7. [8] Chang TC, Wysk RA, Wang HP. Computer-aided manufacturing, Englewood Cli!s, NJ: Prentice-Hall Inc, 1991. [9] Lee KI, Lee JW, Lee JM. Pattern recognition and process planning for prismatic workpieces by knowledge based approach. Ann CIRP 1989;38(1):485}8. [10] Shpitalni M, Fischer A. CSG representation as a basis for extraction of machining features. Ann CIRP 1991;40(1):157}60. [11] Kumara SRT, Kao CY, Gallagher MG, Kasturi R. 3-D interacting manufacturing feature representation. Ann CIRP 1994; 43(1):133}6. [12] Venuvinod PK, Yuen CF. E$cient automated geometric feature recognition through feature coding. Ann CIRP 1994;43(1):413}6. [13] Giusti F, Santochi M, Dini G. COATS: an expert module for optimal tool selection. Ann CIRP 1986;35(1):337}40. [14] Mathieu L, Bourdet P. Tool automatic choice: a step to elaborate automatically process planning. Ann CIRP 1987;36(1):347}50. [15] Rho HM, Geelink R, van't Erve AH, Kals HJJ. An integrated cutting tool selection and operation sequencing method. Ann CIRP 1992;41(1):517}20. [16] Hinduja H, Barrow G. SITS * A semi-intelligent tool selection system for turned components. Ann CIRP 1993;42(1):535}9. [17] Rao AS, George! MP. Modeling rational agents within a BDIarchitecture. Proceedings of Knowledge Representation and Reasoning. 1991. p. 473}84. [18] IEC Technical Committee, Function Blocks for Industrial-Process Measurement and Control Systems, Part 1: Architecture. IEC-TC65/WG6 Committee Draft, June, 1997. [19] Lewis R. Design of distributed control systems in the next millennium. Comput Control Engng J, August 1997;148}52. [20] Wang L, Balasubramanian S, Norrie DH, Brennan RW. Agentbased control system for next generation manufacturing. Proceedings of IEEE ISIC/CIRA/ISAS Joint conference on Science and Technology of Intelligent Systems. September, 1998. p. 78}83. [21] Wang L. An approach to collaborative design and intelligent manufacturing. Proceedings of SCI'99/ ISAS'99 August, 1999. Vol.7, p. 431}37.